Unblocked vs Sourcegraph Cody: Which Gives Coding Agents Better Context? (2026)

Disclosure: Unblocked is one of the two products compared here. Sourcegraph is represented using its own published documentation and product materials. The goal is an honest comparison, not a neutral one.

Updated September 10, 2026 for Sourcegraph's Cody plan changes, the Amp spin-out, Sourcegraph's MCP server, and current Unblocked pricing, sources, and customer results.

Unblocked gives coding agents better context than Sourcegraph Cody because it delivers the decisions behind the code, from PRs, Slack, Jira, Confluence, Notion, and production systems, alongside the code itself, and Cloudbeds measured the result at 66% fewer tokens per agent task (Cloudbeds customer story). Cody indexes code and does that well. The trouble is that in 2026 the agent already has the code. Claude Code, Cursor, and Codex open the repository, grep it, and read every file they need; a search index makes that faster, and speed is not what produces the almost-right pull request. A field study of 78 professional developers found they spend 58% of their working time on program comprehension (Xia et al., IEEE TSE, 2018), and much of what they are trying to comprehend never made it into the repository: why a pattern was chosen, what was tried and reverted, who owns the decision, what the last incident taught the team. An engineering manager at Cloudbeds put the limit of the code in one line:

Your code tells you what's happening today. It doesn't tell you why. And to make good decisions, why is what matters. That's what Unblocked gives you.

Charles Thompson — Engineering Manager, Reservation and Guest, Cloudbeds

That is where code search and code context diverge, and it is the core of the Unblocked vs Sourcegraph Cody question.

For foundational context, see our post on what a context engine is.

---

Table of Contents

1. What Are You Actually Comparing?
2. What Does Sourcegraph Cody Do Well?
3. Where Does Sourcegraph Cody Stop?
4. What Does Unblocked Add Beyond Code Search?
5. How Do the Two Compare on Key Dimensions?
6. Which One Fits Your Team?
7. FAQ

---

What Are You Actually Comparing?#

Atlassian's 2025 State of Developer Experience report found that 50% of developers lose ten or more hours a week to organizational inefficiencies, and that finding information (services, docs, APIs) ranks as the top time-waster (Atlassian, 2025). Sourcegraph Cody addresses the code-finding portion of that lost time; Unblocked addresses the decision-context portion that code search misses.

Sourcegraph Cody is an AI coding assistant built on Sourcegraph's code search engine. Per Sourcegraph's documentation, it runs in VS Code, JetBrains, Visual Studio, the Sourcegraph web app, and a CLI, and it uses Sourcegraph's Search API to pull context from local and remote repositories (Sourcegraph docs). Since July 23, 2025, Cody ships only as part of Sourcegraph Enterprise. Sourcegraph discontinued Cody Free and Cody Pro on that date, removed Cody from its Enterprise Starter plan, and directed individual developers to Amp, its agentic coding tool (Sourcegraph, June 2025). Amp then spun out as an independent company, Amp Inc., in December 2025, leaving Sourcegraph focused on code search and code understanding for large codebases (Sourcegraph, December 2025). Sourcegraph has also shipped its own MCP server, which exposes code search, navigation, history, and Deep Search to Claude Code, Cursor, Codex, Copilot, and Amp on Enterprise plans (Sourcegraph docs).

Unblocked is the context engine for agentic software development. It sits underneath whichever agent you already use and connects to code, PRs, Slack, Jira, Linear, Confluence, Notion, Google Drive, CI, and production systems such as Datadog and Sentry, then synthesizes the institutional knowledge around the code. It delivers that context through MCP to Claude Code, Cursor, GitHub Copilot, and Codex, and also through a CLI, IDE integrations, Slack and Teams, an API, and AI code review on pull requests (getunblocked.com/context-engine).

The distinction matters because they solve different failure modes. Cody helps when the problem is "I can't find the code." That was the hard problem when Sourcegraph started; today a coding agent finds the code in seconds without any index at all. Unblocked helps with the problem agents actually fail on: "I found the code but I don't know why it's written this way, what's been tried before, or what work is already in progress." Search returns candidates. Agents need a decision.

For a related comparison, see context engine vs enterprise search.

---

What Does Sourcegraph Cody Do Well?#

Sourcegraph has been building code search for more than a decade, and its pricing page positions the Enterprise plan for "the world's most complex codebases," with a public index it says spans more than a million repositories (Sourcegraph pricing). Cody inherits that search infrastructure and adds chat, auto-edit, prompts, and debugging grounded in code, per Sourcegraph's own documentation.

Code navigation at scale#

Cody's biggest advantage is Sourcegraph's code graph. Cross-repository code intelligence, precise go-to-definition, find-references, and symbol search are genuinely difficult at enterprise scale. Teams running thousands of repositories across multiple code hosts benefit from having a single search layer that works regardless of where the code lives. Sourcegraph's newer Deep Search, which it describes as an agentic natural-language search that aggregates context across mono-repo and multi-repo codebases, extends that reach to agents (Sourcegraph, December 2025). Few products do this as well as Sourcegraph does.

AI grounded in code#

Cody's chat and autocomplete use the code graph as context. When you ask Cody a question, it retrieves relevant code snippets and files from your indexed repositories. This is meaningfully better than a generic LLM that has never seen your codebase. For questions like "where is the payment retry logic?" or "show me all callers of this function," Cody is fast and accurate. Its agentic context fetching can also call the terminal, the web, and local MCP servers once an Enterprise admin enables the relevant feature flags (Sourcegraph docs).

IDE integration#

Cody ships extensions for VS Code, JetBrains, and Visual Studio, plus a web client and CLI, with inline chat, autocomplete, and edit commands. The IDE experience is polished for individual engineers who want to search and ask questions without leaving their editor. Sourcegraph's 2026 changelog shows Cody still receiving model updates, including Claude Opus 4.7 in April and the GPT-5.6 family in August (Sourcegraph changelog, April 2026; Sourcegraph changelog, August 2026).

In our own evaluation, Sourcegraph's cross-repository search consistently returned relevant code matches faster than grep-based alternatives for organizations with 500+ repositories.

But here's the question worth asking. How often is the actual blocker "I can't find the code"? And how often is it "I found the code, but I'm missing something that isn't in the code at all"?

---

Where Does Sourcegraph Cody Stop?#

GitHub's 2025 Octoverse report counted 518.7 million merged pull requests, up 29% year over year, and found that 80% of new developers on GitHub use Copilot in their first week (GitHub, 2025). Code is arriving faster than ever, yet 66% of developers still name "almost right" AI output as their top frustration (Stack Overflow, 2025). The gap between almost right and mergeable is where Sourcegraph Cody's scope ends.

The boundary is the code itself#

Cody reasons over code. It does not ingest Slack conversations where your team debated an API design. It doesn't read Jira tickets tracking an ongoing migration. It doesn't parse PR review threads where a senior engineer explained why a pattern was deprecated. It doesn't index Confluence pages documenting your team's error-handling conventions.

Here is what that looks like on a real task. Ask code search which service owns the upload retry policy and it finds the service and the values in seconds. It cannot tell the agent about the incident that set those values, the Slack argument where a shorter backoff was rejected, the PR where someone tried to remove the retry and reverted it a week later, or who made the call. So the agent "improves" the retry logic, the PR looks clean, and a human unwinds it after the pager goes off. At Webflow, an engineer asked an agent to implement a feature and Unblocked found the PR that had already implemented it elsewhere in the codebase, short-circuiting the whole task (Webflow customer story). Code search could have shown the agent every file. It could not have told it the work was done.

Cody's agentic context fetching can reach outside the repository through MCP, but per Sourcegraph's documentation the capability is disabled by default, gated behind an Enterprise feature flag, and limited to local MCP servers (Sourcegraph docs). Each server it reaches exposes one raw system. Sourcegraph's own MCP server, likewise, serves the code graph. Those non-code sources aren't edge cases. They're where most institutional knowledge lives.

Search finds the first plausible match, and agents act on it#

Agents are subject to a well-documented bias we call satisfaction of search: keep searching until the first plausible answer appears, then act on it. A code index makes that failure faster, because the first plausible match is always a file. Unblocked traverses the whole knowledge graph in one pass and hands the agent the best answer for the task rather than the first hit (getunblocked.com/context-engine). The cost of search-and-hope is measurable. When Cloudbeds ran the same incident investigation both ways, the search-driven path spent about fifteen thousand tokens assembling a picture that Unblocked returned, already reconciled, for about five thousand (Cloudbeds customer story). Advidi tried wiring native connectors to its agents first and got "extreme context bloat," loading whole wikis to find one decision, before consolidating on Unblocked (Advidi customer story).

The code is often wrong about the present#

A code index is a faithful record of what shipped, and what shipped is frequently behind what the team has decided. The migration that Jira says is half done, the convention that changed in a review thread last week, the endpoint that Slack agreed to deprecate on Friday: none of it is in the repository yet, and an agent grounded only in the code will confidently build against the past. An engineering manager at Clio described the effect directly:

LLMs are only as good as their context, and code alone gives them a narrow view — the codebase is always behind where the team wants to go. When I plugged Unblocked into our context-gathering and PR review steps, it brought in the Slack conversations where real architectural decisions get made. We went from three rounds of PR review to one before the code was production-ready.

Pablo Vallejo — Engineering Manager, Clio

No cross-source conflict resolution#

When your Notion doc says one thing, the code says another, and a Slack thread from last month says a third, your AI agent needs help deciding which source to trust. Cody doesn't see the conflict because it only sees the code. The agent produces code that compiles but violates a convention documented elsewhere. Unblocked weighs recency, authority, and proximity to the developer's own work to settle that conflict before the agent ever sees it, and ships the answer with links to the sources it chose (getunblocked.com/context-engine).

Assistant first, code-graph MCP second#

Cody is an assistant. If your team runs Claude Code, Cursor, Copilot, or Codex, Cody itself doesn't plug into those tools. Sourcegraph's separate MCP server does, on Enterprise plans, but what it carries is search, navigation, history, and Deep Search over code (Sourcegraph docs). The agent still arrives at every PR thread, ticket, and design debate as a stranger, and it arrives faster.

Across the enterprise evaluations we've been part of, the pattern is consistent: code search covers the smaller share of the context problem, and it is the share the agent could already cover on its own. The larger share is scattered across Slack, Jira, PR discussions, and docs, contradicts itself, and changes weekly. Solving the code share is valuable. Stopping there leaves the part that produces wrong pull requests untouched.

Pricing and licensing shifts#

Sourcegraph's lineup has changed substantially since this comparison was first published. On June 25, 2025, Sourcegraph announced that Cody Free and Cody Pro would end on July 23, 2025, that Enterprise Starter would no longer include Cody, and that individuals should move to Amp (Sourcegraph, June 2025). Sourcegraph's pricing page now lists a single Enterprise plan "starting at $16K" as a minimum annual contract, with credits for AI features that scale with team size (Sourcegraph pricing). Amp, now a separate company, sells $20 and $200 monthly tiers plus usage-based billing and an enterprise option (Amp pricing). Engineering leaders evaluating a Sourcegraph alternative often cite that churn alongside the feature gaps above.

---

What Does Unblocked Add Beyond Code Search?#

The Pragmatic Engineer's 2026 survey of nearly 900 engineers found that 95% use AI tools at least weekly and 63.5% of staff-plus engineers run agents regularly, while Anthropic's 2026 Agentic Coding Trends Report found developers use AI in roughly 60% of their work but can "fully delegate" only 0-20% of tasks (Pragmatic Engineer, 2026; Anthropic, 2026). That delegation gap is a context gap, and closing it is what Unblocked was built for.

The sources that matter most#

Unblocked ingests the full engineering surface: GitHub, GitLab, Bitbucket, and Azure DevOps (code and PR history), Jira, Linear, Asana, and GitHub Issues (tickets and epics), Slack and Microsoft Teams (conversations and decisions), Confluence, Notion, Google Drive, SharePoint, Coda, and Stack Overflow for Teams (documentation), CI systems including GitHub Actions, Buildkite, CircleCI, and Jenkins, and production signals from Datadog and Sentry (getunblocked.com/pricing). Code is one first-class input among many.

One reconciled answer, scored and compressed#

Long-context research explains why handing an agent raw output from a dozen systems fails. The NoLiMa benchmark found that at 32K tokens, 11 of 13 models fell below 50% of their short-context baseline once the answer required more than literal keyword matching (Modarressi et al., 2025). A follow-up across more than 150,000 controlled runs found that the smaller the relevant passage relative to the surrounding text, the harder models find it (Bianchi et al., 2025). Anthropic's engineering team describes the goal as "finding the smallest possible set of high-signal tokens that maximize the likelihood of some desired outcome" (Anthropic, 2025).

Unblocked is built around that constraint. It traverses a single knowledge graph across every source in one pass, resolves contradictions by recency, authority, and proximity, personalizes relevance to the repos and systems each developer actually works in, and hands the agent context that is scored, compressed, and assembled server-side. One MCP connection replaces the per-system connectors that each add overhead before the agent does any real work (getunblocked.com/context-engine).

Agent-agnostic delivery via MCP#

Unblocked doesn't ask you to switch agents. Claude Code, Cursor, GitHub Copilot, and Codex all get the same decision-grade context through one MCP connection, and a CLI, IDE integrations, Slack and Teams apps, an API, and AI code review on GitHub PRs and GitLab MRs round out the surface. Webflow's engineering productivity org runs remote agentic development at scale with Claude, Cursor, and Codex on top of Unblocked, which engineers there rated a top-three engineering tool with an 82 CSAT in the quarterly tool survey (Webflow customer story). Your team keeps its preferred tools; Unblocked is the infrastructure feeding all of them.

For more on the discipline behind this, see the context engineering guide.

Permission enforcement end to end#

Unblocked inherits permissions from each connected source system, per user. An agent acting on behalf of a junior engineer sees only what that engineer is authorized to see, with a full audit trail. Data Shield enforces this at ingestion and at delivery. SOC 2 Type II, CASA Tier II, SAML SSO with SCIM provisioning, RBAC, and audit logs are built in (getunblocked.com/security).

What teams running agents on it report#

Advidi's CTO evaluated the native MCP servers for GitHub, Confluence, and Jira before connecting 400+ repositories and its incident agent to Unblocked instead:

Unblocked combines all of the tools we use into one product, one single MCP. And the search is efficient: the context it gathers is relevant and recent. I haven't found another tool that does that as effectively as Unblocked does.

Jorden van Breemen — CTO, Advidi

At Cloudbeds, where 987 GitHub repositories and more than a decade of decisions sit behind the product, an engineering manager benchmarked the same question with and without Unblocked:

Unblocked used about a third of the tokens. Five thousand versus fifteen thousand. And the thing I thought was amazing is that Unblocked had the answer already. The rest was just my AI dressing it up. After that I opened up about twenty PRs, and now the first thing all of our skills do is ask Unblocked.

Charles Thompson — Engineering Manager, Reservation and Guest, Cloudbeds

Both reflect a pattern we hear consistently from senior engineers: the value is in the decisions around the code. The Slack thread where a design was debated. The Jira epic tracking a migration. The PR review that explains why the deprecated client is still in production.

In a controlled internal test, the same agent on the same codebase completed the same task with 48% fewer tokens and 83% faster with Unblocked feeding context upstream. A single-task benchmark, but consistent with what Cloudbeds measured after deployment (getunblocked.com).

---

How Do the Two Compare on Key Dimensions?#

Gartner projects that 40% of enterprise applications will be integrated with task-specific AI agents by the end of 2026, up from under 5% in 2025 (Gartner, 2025). As agent adoption scales, context infrastructure that feeds any agent becomes more valuable than any single agent's proprietary features.

Sourcegraph Cody vs Unblocked comparison matrix:

Code search and code context are complementary capabilities. Source: Sourcegraph pricing and documentation as of September 2026; Unblocked pricing, documentation, and customer stories.

The table reveals the core difference. Cody is deep on code and narrow on everything else, while Unblocked treats code as one input among many and does the reconciliation work between sources itself. For a team whose agents already read the repository, the first column describes a capability they have and the second describes the one they are missing.

---

Which One Fits Your Team?#

DORA's 2026 report on the ROI of AI-assisted software development argues that the greatest returns come from the organizational system around the tools: the quality of the internal platform, the clarity of workflows, and the alignment of teams, rather than from the tools themselves (DORA, 2026). Context is part of that system, and the tooling question shows up every time an engineer or an agent asks, "Why is this code written this way?"

Pick Sourcegraph Cody when:#

• Your primary pain point is finding code across a large, multi-repository codebase
• Your team needs cross-repository code intelligence (go-to-definition, find-references) at scale
• You are already a Sourcegraph Enterprise customer and want an AI assistant that answers code-level questions in the IDE
• Your institutional knowledge lives almost entirely in the code itself, and you accept that your agents will keep meeting every PR thread, ticket, and design decision as strangers

Pick Unblocked when:#

• Your AI agents already have the code open in front of them and still produce pull requests that miss conventions, deprecated patterns, or in-flight decisions
• Your institutional knowledge is scattered across Slack, Jira, Confluence, PR discussions, and docs
• Your team uses multiple AI agents (Claude Code, Cursor, Copilot, Codex) and you want one context engine behind all of them
• You need permission-enforced context delivery with SOC 2, SAML SSO, and audit trails
• You want to spend fewer tokens and less time babysitting agents, the outcome Cloudbeds and Webflow report

Consider both when:#

Can you run Sourcegraph and Unblocked together? Yes. They solve different problems. Sourcegraph handles code search and navigation, and its MCP server can sit beside Unblocked's in the same agent configuration. Unblocked handles the institutional context that code search doesn't cover. Some enterprise teams run Sourcegraph for code intelligence and Unblocked for the decision-grade context that feeds their AI agents. The combination is additive.

The most common mistake we see in Sourcegraph alternative evaluations is treating code context as a one-dimensional problem. Teams compare Cody to Unblocked on code search alone and miss that Unblocked is solving a different problem entirely. The question isn't "which searches code better?" It's "where does your team's context actually live?"

Read the full context engineering guide for the discipline above the tools.

---

FAQ#

Is Sourcegraph Cody a free alternative to Unblocked?#

No, on both counts. Sourcegraph ended Cody Free and Cody Pro on July 23, 2025, so Cody is available only with a Sourcegraph Enterprise contract (Sourcegraph, June 2025). And the two products solve different problems: Cody provides AI chat grounded in code search, while Unblocked provides cross-source context from Slack, Jira, PR discussions, and docs, with a 21-day trial at $29 per user per month (getunblocked.com/pricing). Stack Overflow's 2025 survey found 84% of developers use or plan to use AI tools, but 66% still cite "almost right" answers as their top frustration, which suggests code grounding alone isn't sufficient (Stack Overflow, 2025).

What happened to Cody Free and Cody Pro, and is Amp the replacement?#

Sourcegraph stopped new Cody Free and Pro signups on June 25, 2025 and discontinued both plans on July 23, 2025, offering Amp credits to affected users and describing Amp as "the best path forward" for them (Sourcegraph, June 2025). In December 2025, Sourcegraph and Amp became separate companies, with Amp Inc. focused on coding agents and Sourcegraph on code search (Sourcegraph, December 2025). Amp is a coding agent, so it competes with Claude Code and Cursor rather than with Unblocked, and it can consume Unblocked's context over MCP like any other agent.

Can I use Unblocked with my existing Sourcegraph setup?#

Yes. Unblocked connects to the same code hosts Sourcegraph indexes (GitHub, GitLab, Bitbucket, Azure DevOps) and adds the non-code sources that Sourcegraph doesn't cover. The two products don't conflict. Your engineers keep Sourcegraph for code search, and their agents can hold both Sourcegraph's MCP server and Unblocked's in the same configuration.

Does Sourcegraph's MCP server make Unblocked unnecessary?#

No. Sourcegraph's MCP server exposes code search, navigation, history, and Deep Search from your Sourcegraph instance to agents such as Claude Code, Cursor, and Codex (Sourcegraph docs). It gives an agent a better view of the code. Unblocked gives the agent the rest of the context: the PR threads, Slack decisions, tickets, docs, and production signals, reconciled into one answer and scoped to each user's permissions. Advidi's CTO tried the per-system MCP route first and found it produced "extreme context bloat" before consolidating on Unblocked (Advidi customer story).

Does Unblocked do code search?#

Unblocked ingests code repositories, but its primary value isn't symbol-level code search. It's the synthesis of code with the conversations, tickets, reviews, and documentation that explain the code. For pure code navigation, Sourcegraph's infrastructure is purpose-built. For understanding why the code is the way it is, Unblocked provides the broader picture.

What if my team standardizes on Cody as our AI assistant?#

If your team commits to Cody as the primary AI assistant, Cody handles code-grounded chat and autocomplete inside the IDE. The question is whether your engineers and their agents also need context from Slack, Jira, and PR discussions. If they do, Unblocked fills that gap and delivers context through MCP, its web app, Slack and Teams apps, CLI, API, and code review. DORA's 2026 ROI report makes the broader point: AI returns depend on the organizational system around the tools, including how clearly knowledge flows through it (DORA, 2026).

---

Code Search vs Code Understanding#

The distinction between these two products maps to a distinction in how engineering teams actually work. Code search answers "where is it?" Code understanding answers "why is it this way, and what should I know before I change it?"

Sourcegraph built the best code search platform in the industry, and its MCP server now hands that search to agents. That's a genuine achievement. It also solves a problem that coding agents have largely stopped having. An agent with a repository checked out can find any symbol; what it cannot find is the reason the symbol exists, the review that rejected the alternative, or the incident that set the number it is about to change. The context problem lives in the Slack threads, Jira epics, PR discussions, and documentation that shape every engineering decision, and a code index does not reach any of them.

Unblocked was built for that extended surface. It covers what code search can't reach, reconciles it, scopes it to each user's permissions, and delivers it to whichever agent your team already uses, without replacing the search itself.

The right choice depends on your bottleneck. If your engineers can't find code, solve that first. If they find code but keep producing "almost right" output because they're missing the institutional context around it, that's the problem Unblocked closes. Advidi's CTO evaluated the alternatives and hasn't found a tool that does context as effectively; Cloudbeds now makes Unblocked the first step of every agent skill; Webflow calls it the brain that makes its agents work. That is what it looks like to stop babysitting your agents.

For a deeper exploration, read about what a context engine is and how it differs from search.

Continue reading:

• What is a context engine? - the architectural view of engineering context infrastructure.
• Context engine vs enterprise search - where context engines differ from search.
• The context engineering guide - the discipline above the tools.

---

Brandon Waselnuk is a content lead at Unblocked.